Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / scsi / sd.c
blob679c2c02504763ba31a21c425d58bdfca02baba8
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
36 #include <linux/module.h>
37 #include <linux/fs.h>
38 #include <linux/kernel.h>
39 #include <linux/mm.h>
40 #include <linux/bio.h>
41 #include <linux/genhd.h>
42 #include <linux/hdreg.h>
43 #include <linux/errno.h>
44 #include <linux/idr.h>
45 #include <linux/interrupt.h>
46 #include <linux/init.h>
47 #include <linux/blkdev.h>
48 #include <linux/blkpg.h>
49 #include <linux/blk-pm.h>
50 #include <linux/delay.h>
51 #include <linux/mutex.h>
52 #include <linux/string_helpers.h>
53 #include <linux/async.h>
54 #include <linux/slab.h>
55 #include <linux/sed-opal.h>
56 #include <linux/pm_runtime.h>
57 #include <linux/pr.h>
58 #include <linux/t10-pi.h>
59 #include <linux/uaccess.h>
60 #include <asm/unaligned.h>
62 #include <scsi/scsi.h>
63 #include <scsi/scsi_cmnd.h>
64 #include <scsi/scsi_dbg.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_driver.h>
67 #include <scsi/scsi_eh.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi_ioctl.h>
70 #include <scsi/scsicam.h>
72 #include "sd.h"
73 #include "scsi_priv.h"
74 #include "scsi_logging.h"
76 MODULE_AUTHOR("Eric Youngdale");
77 MODULE_DESCRIPTION("SCSI disk (sd) driver");
78 MODULE_LICENSE("GPL");
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
101 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
102 #define SD_MINORS 16
103 #else
104 #define SD_MINORS 0
105 #endif
107 static void sd_config_discard(struct scsi_disk *, unsigned int);
108 static void sd_config_write_same(struct scsi_disk *);
109 static int sd_revalidate_disk(struct gendisk *);
110 static void sd_unlock_native_capacity(struct gendisk *disk);
111 static int sd_probe(struct device *);
112 static int sd_remove(struct device *);
113 static void sd_shutdown(struct device *);
114 static int sd_suspend_system(struct device *);
115 static int sd_suspend_runtime(struct device *);
116 static int sd_resume(struct device *);
117 static void sd_rescan(struct device *);
118 static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120 static int sd_done(struct scsi_cmnd *);
121 static void sd_eh_reset(struct scsi_cmnd *);
122 static int sd_eh_action(struct scsi_cmnd *, int);
123 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124 static void scsi_disk_release(struct device *cdev);
126 static DEFINE_IDA(sd_index_ida);
128 /* This semaphore is used to mediate the 0->1 reference get in the
129 * face of object destruction (i.e. we can't allow a get on an
130 * object after last put) */
131 static DEFINE_MUTEX(sd_ref_mutex);
133 static struct kmem_cache *sd_cdb_cache;
134 static mempool_t *sd_cdb_pool;
135 static mempool_t *sd_page_pool;
137 static const char *sd_cache_types[] = {
138 "write through", "none", "write back",
139 "write back, no read (daft)"
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 bool wc = false, fua = false;
146 if (sdkp->WCE) {
147 wc = true;
148 if (sdkp->DPOFUA)
149 fua = true;
152 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
155 static ssize_t
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 const char *buf, size_t count)
159 int ct, rcd, wce, sp;
160 struct scsi_disk *sdkp = to_scsi_disk(dev);
161 struct scsi_device *sdp = sdkp->device;
162 char buffer[64];
163 char *buffer_data;
164 struct scsi_mode_data data;
165 struct scsi_sense_hdr sshdr;
166 static const char temp[] = "temporary ";
167 int len;
169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 /* no cache control on RBC devices; theoretically they
171 * can do it, but there's probably so many exceptions
172 * it's not worth the risk */
173 return -EINVAL;
175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 buf += sizeof(temp) - 1;
177 sdkp->cache_override = 1;
178 } else {
179 sdkp->cache_override = 0;
182 ct = sysfs_match_string(sd_cache_types, buf);
183 if (ct < 0)
184 return -EINVAL;
186 rcd = ct & 0x01 ? 1 : 0;
187 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
189 if (sdkp->cache_override) {
190 sdkp->WCE = wce;
191 sdkp->RCD = rcd;
192 sd_set_flush_flag(sdkp);
193 return count;
196 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 sdkp->max_retries, &data, NULL))
198 return -EINVAL;
199 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 data.block_descriptor_length);
201 buffer_data = buffer + data.header_length +
202 data.block_descriptor_length;
203 buffer_data[2] &= ~0x05;
204 buffer_data[2] |= wce << 2 | rcd;
205 sp = buffer_data[0] & 0x80 ? 1 : 0;
206 buffer_data[0] &= ~0x80;
209 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
210 * received mode parameter buffer before doing MODE SELECT.
212 data.device_specific = 0;
214 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
215 sdkp->max_retries, &data, &sshdr)) {
216 if (scsi_sense_valid(&sshdr))
217 sd_print_sense_hdr(sdkp, &sshdr);
218 return -EINVAL;
220 sd_revalidate_disk(sdkp->disk);
221 return count;
224 static ssize_t
225 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
226 char *buf)
228 struct scsi_disk *sdkp = to_scsi_disk(dev);
229 struct scsi_device *sdp = sdkp->device;
231 return sprintf(buf, "%u\n", sdp->manage_start_stop);
234 static ssize_t
235 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
236 const char *buf, size_t count)
238 struct scsi_disk *sdkp = to_scsi_disk(dev);
239 struct scsi_device *sdp = sdkp->device;
240 bool v;
242 if (!capable(CAP_SYS_ADMIN))
243 return -EACCES;
245 if (kstrtobool(buf, &v))
246 return -EINVAL;
248 sdp->manage_start_stop = v;
250 return count;
252 static DEVICE_ATTR_RW(manage_start_stop);
254 static ssize_t
255 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
257 struct scsi_disk *sdkp = to_scsi_disk(dev);
259 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
262 static ssize_t
263 allow_restart_store(struct device *dev, struct device_attribute *attr,
264 const char *buf, size_t count)
266 bool v;
267 struct scsi_disk *sdkp = to_scsi_disk(dev);
268 struct scsi_device *sdp = sdkp->device;
270 if (!capable(CAP_SYS_ADMIN))
271 return -EACCES;
273 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
274 return -EINVAL;
276 if (kstrtobool(buf, &v))
277 return -EINVAL;
279 sdp->allow_restart = v;
281 return count;
283 static DEVICE_ATTR_RW(allow_restart);
285 static ssize_t
286 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
288 struct scsi_disk *sdkp = to_scsi_disk(dev);
289 int ct = sdkp->RCD + 2*sdkp->WCE;
291 return sprintf(buf, "%s\n", sd_cache_types[ct]);
293 static DEVICE_ATTR_RW(cache_type);
295 static ssize_t
296 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
298 struct scsi_disk *sdkp = to_scsi_disk(dev);
300 return sprintf(buf, "%u\n", sdkp->DPOFUA);
302 static DEVICE_ATTR_RO(FUA);
304 static ssize_t
305 protection_type_show(struct device *dev, struct device_attribute *attr,
306 char *buf)
308 struct scsi_disk *sdkp = to_scsi_disk(dev);
310 return sprintf(buf, "%u\n", sdkp->protection_type);
313 static ssize_t
314 protection_type_store(struct device *dev, struct device_attribute *attr,
315 const char *buf, size_t count)
317 struct scsi_disk *sdkp = to_scsi_disk(dev);
318 unsigned int val;
319 int err;
321 if (!capable(CAP_SYS_ADMIN))
322 return -EACCES;
324 err = kstrtouint(buf, 10, &val);
326 if (err)
327 return err;
329 if (val <= T10_PI_TYPE3_PROTECTION)
330 sdkp->protection_type = val;
332 return count;
334 static DEVICE_ATTR_RW(protection_type);
336 static ssize_t
337 protection_mode_show(struct device *dev, struct device_attribute *attr,
338 char *buf)
340 struct scsi_disk *sdkp = to_scsi_disk(dev);
341 struct scsi_device *sdp = sdkp->device;
342 unsigned int dif, dix;
344 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
345 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
347 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
348 dif = 0;
349 dix = 1;
352 if (!dif && !dix)
353 return sprintf(buf, "none\n");
355 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
357 static DEVICE_ATTR_RO(protection_mode);
359 static ssize_t
360 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
362 struct scsi_disk *sdkp = to_scsi_disk(dev);
364 return sprintf(buf, "%u\n", sdkp->ATO);
366 static DEVICE_ATTR_RO(app_tag_own);
368 static ssize_t
369 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
370 char *buf)
372 struct scsi_disk *sdkp = to_scsi_disk(dev);
374 return sprintf(buf, "%u\n", sdkp->lbpme);
376 static DEVICE_ATTR_RO(thin_provisioning);
378 /* sysfs_match_string() requires dense arrays */
379 static const char *lbp_mode[] = {
380 [SD_LBP_FULL] = "full",
381 [SD_LBP_UNMAP] = "unmap",
382 [SD_LBP_WS16] = "writesame_16",
383 [SD_LBP_WS10] = "writesame_10",
384 [SD_LBP_ZERO] = "writesame_zero",
385 [SD_LBP_DISABLE] = "disabled",
388 static ssize_t
389 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
390 char *buf)
392 struct scsi_disk *sdkp = to_scsi_disk(dev);
394 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
397 static ssize_t
398 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
399 const char *buf, size_t count)
401 struct scsi_disk *sdkp = to_scsi_disk(dev);
402 struct scsi_device *sdp = sdkp->device;
403 int mode;
405 if (!capable(CAP_SYS_ADMIN))
406 return -EACCES;
408 if (sd_is_zoned(sdkp)) {
409 sd_config_discard(sdkp, SD_LBP_DISABLE);
410 return count;
413 if (sdp->type != TYPE_DISK)
414 return -EINVAL;
416 mode = sysfs_match_string(lbp_mode, buf);
417 if (mode < 0)
418 return -EINVAL;
420 sd_config_discard(sdkp, mode);
422 return count;
424 static DEVICE_ATTR_RW(provisioning_mode);
426 /* sysfs_match_string() requires dense arrays */
427 static const char *zeroing_mode[] = {
428 [SD_ZERO_WRITE] = "write",
429 [SD_ZERO_WS] = "writesame",
430 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
431 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
434 static ssize_t
435 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
436 char *buf)
438 struct scsi_disk *sdkp = to_scsi_disk(dev);
440 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
443 static ssize_t
444 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
445 const char *buf, size_t count)
447 struct scsi_disk *sdkp = to_scsi_disk(dev);
448 int mode;
450 if (!capable(CAP_SYS_ADMIN))
451 return -EACCES;
453 mode = sysfs_match_string(zeroing_mode, buf);
454 if (mode < 0)
455 return -EINVAL;
457 sdkp->zeroing_mode = mode;
459 return count;
461 static DEVICE_ATTR_RW(zeroing_mode);
463 static ssize_t
464 max_medium_access_timeouts_show(struct device *dev,
465 struct device_attribute *attr, char *buf)
467 struct scsi_disk *sdkp = to_scsi_disk(dev);
469 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
472 static ssize_t
473 max_medium_access_timeouts_store(struct device *dev,
474 struct device_attribute *attr, const char *buf,
475 size_t count)
477 struct scsi_disk *sdkp = to_scsi_disk(dev);
478 int err;
480 if (!capable(CAP_SYS_ADMIN))
481 return -EACCES;
483 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
485 return err ? err : count;
487 static DEVICE_ATTR_RW(max_medium_access_timeouts);
489 static ssize_t
490 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
491 char *buf)
493 struct scsi_disk *sdkp = to_scsi_disk(dev);
495 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
498 static ssize_t
499 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
500 const char *buf, size_t count)
502 struct scsi_disk *sdkp = to_scsi_disk(dev);
503 struct scsi_device *sdp = sdkp->device;
504 unsigned long max;
505 int err;
507 if (!capable(CAP_SYS_ADMIN))
508 return -EACCES;
510 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
511 return -EINVAL;
513 err = kstrtoul(buf, 10, &max);
515 if (err)
516 return err;
518 if (max == 0)
519 sdp->no_write_same = 1;
520 else if (max <= SD_MAX_WS16_BLOCKS) {
521 sdp->no_write_same = 0;
522 sdkp->max_ws_blocks = max;
525 sd_config_write_same(sdkp);
527 return count;
529 static DEVICE_ATTR_RW(max_write_same_blocks);
531 static ssize_t
532 zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
534 struct scsi_disk *sdkp = to_scsi_disk(dev);
536 if (sdkp->device->type == TYPE_ZBC)
537 return sprintf(buf, "host-managed\n");
538 if (sdkp->zoned == 1)
539 return sprintf(buf, "host-aware\n");
540 if (sdkp->zoned == 2)
541 return sprintf(buf, "drive-managed\n");
542 return sprintf(buf, "none\n");
544 static DEVICE_ATTR_RO(zoned_cap);
546 static ssize_t
547 max_retries_store(struct device *dev, struct device_attribute *attr,
548 const char *buf, size_t count)
550 struct scsi_disk *sdkp = to_scsi_disk(dev);
551 struct scsi_device *sdev = sdkp->device;
552 int retries, err;
554 err = kstrtoint(buf, 10, &retries);
555 if (err)
556 return err;
558 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
559 sdkp->max_retries = retries;
560 return count;
563 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
564 SD_MAX_RETRIES);
565 return -EINVAL;
568 static ssize_t
569 max_retries_show(struct device *dev, struct device_attribute *attr,
570 char *buf)
572 struct scsi_disk *sdkp = to_scsi_disk(dev);
574 return sprintf(buf, "%d\n", sdkp->max_retries);
577 static DEVICE_ATTR_RW(max_retries);
579 static struct attribute *sd_disk_attrs[] = {
580 &dev_attr_cache_type.attr,
581 &dev_attr_FUA.attr,
582 &dev_attr_allow_restart.attr,
583 &dev_attr_manage_start_stop.attr,
584 &dev_attr_protection_type.attr,
585 &dev_attr_protection_mode.attr,
586 &dev_attr_app_tag_own.attr,
587 &dev_attr_thin_provisioning.attr,
588 &dev_attr_provisioning_mode.attr,
589 &dev_attr_zeroing_mode.attr,
590 &dev_attr_max_write_same_blocks.attr,
591 &dev_attr_max_medium_access_timeouts.attr,
592 &dev_attr_zoned_cap.attr,
593 &dev_attr_max_retries.attr,
594 NULL,
596 ATTRIBUTE_GROUPS(sd_disk);
598 static struct class sd_disk_class = {
599 .name = "scsi_disk",
600 .owner = THIS_MODULE,
601 .dev_release = scsi_disk_release,
602 .dev_groups = sd_disk_groups,
605 static const struct dev_pm_ops sd_pm_ops = {
606 .suspend = sd_suspend_system,
607 .resume = sd_resume,
608 .poweroff = sd_suspend_system,
609 .restore = sd_resume,
610 .runtime_suspend = sd_suspend_runtime,
611 .runtime_resume = sd_resume,
614 static struct scsi_driver sd_template = {
615 .gendrv = {
616 .name = "sd",
617 .owner = THIS_MODULE,
618 .probe = sd_probe,
619 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
620 .remove = sd_remove,
621 .shutdown = sd_shutdown,
622 .pm = &sd_pm_ops,
624 .rescan = sd_rescan,
625 .init_command = sd_init_command,
626 .uninit_command = sd_uninit_command,
627 .done = sd_done,
628 .eh_action = sd_eh_action,
629 .eh_reset = sd_eh_reset,
633 * Don't request a new module, as that could deadlock in multipath
634 * environment.
636 static void sd_default_probe(dev_t devt)
641 * Device no to disk mapping:
643 * major disc2 disc p1
644 * |............|.............|....|....| <- dev_t
645 * 31 20 19 8 7 4 3 0
647 * Inside a major, we have 16k disks, however mapped non-
648 * contiguously. The first 16 disks are for major0, the next
649 * ones with major1, ... Disk 256 is for major0 again, disk 272
650 * for major1, ...
651 * As we stay compatible with our numbering scheme, we can reuse
652 * the well-know SCSI majors 8, 65--71, 136--143.
654 static int sd_major(int major_idx)
656 switch (major_idx) {
657 case 0:
658 return SCSI_DISK0_MAJOR;
659 case 1 ... 7:
660 return SCSI_DISK1_MAJOR + major_idx - 1;
661 case 8 ... 15:
662 return SCSI_DISK8_MAJOR + major_idx - 8;
663 default:
664 BUG();
665 return 0; /* shut up gcc */
669 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
671 struct scsi_disk *sdkp = NULL;
673 mutex_lock(&sd_ref_mutex);
675 if (disk->private_data) {
676 sdkp = scsi_disk(disk);
677 if (scsi_device_get(sdkp->device) == 0)
678 get_device(&sdkp->dev);
679 else
680 sdkp = NULL;
682 mutex_unlock(&sd_ref_mutex);
683 return sdkp;
686 static void scsi_disk_put(struct scsi_disk *sdkp)
688 struct scsi_device *sdev = sdkp->device;
690 mutex_lock(&sd_ref_mutex);
691 put_device(&sdkp->dev);
692 scsi_device_put(sdev);
693 mutex_unlock(&sd_ref_mutex);
696 #ifdef CONFIG_BLK_SED_OPAL
697 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
698 size_t len, bool send)
700 struct scsi_disk *sdkp = data;
701 struct scsi_device *sdev = sdkp->device;
702 u8 cdb[12] = { 0, };
703 int ret;
705 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
706 cdb[1] = secp;
707 put_unaligned_be16(spsp, &cdb[2]);
708 put_unaligned_be32(len, &cdb[6]);
710 ret = scsi_execute_req(sdev, cdb,
711 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
712 buffer, len, NULL, SD_TIMEOUT, sdkp->max_retries, NULL);
713 return ret <= 0 ? ret : -EIO;
715 #endif /* CONFIG_BLK_SED_OPAL */
718 * Look up the DIX operation based on whether the command is read or
719 * write and whether dix and dif are enabled.
721 static unsigned int sd_prot_op(bool write, bool dix, bool dif)
723 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
724 static const unsigned int ops[] = { /* wrt dix dif */
725 SCSI_PROT_NORMAL, /* 0 0 0 */
726 SCSI_PROT_READ_STRIP, /* 0 0 1 */
727 SCSI_PROT_READ_INSERT, /* 0 1 0 */
728 SCSI_PROT_READ_PASS, /* 0 1 1 */
729 SCSI_PROT_NORMAL, /* 1 0 0 */
730 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
731 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
732 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
735 return ops[write << 2 | dix << 1 | dif];
739 * Returns a mask of the protection flags that are valid for a given DIX
740 * operation.
742 static unsigned int sd_prot_flag_mask(unsigned int prot_op)
744 static const unsigned int flag_mask[] = {
745 [SCSI_PROT_NORMAL] = 0,
747 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
748 SCSI_PROT_GUARD_CHECK |
749 SCSI_PROT_REF_CHECK |
750 SCSI_PROT_REF_INCREMENT,
752 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
753 SCSI_PROT_IP_CHECKSUM,
755 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
756 SCSI_PROT_GUARD_CHECK |
757 SCSI_PROT_REF_CHECK |
758 SCSI_PROT_REF_INCREMENT |
759 SCSI_PROT_IP_CHECKSUM,
761 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
762 SCSI_PROT_REF_INCREMENT,
764 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
765 SCSI_PROT_REF_CHECK |
766 SCSI_PROT_REF_INCREMENT |
767 SCSI_PROT_IP_CHECKSUM,
769 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
770 SCSI_PROT_GUARD_CHECK |
771 SCSI_PROT_REF_CHECK |
772 SCSI_PROT_REF_INCREMENT |
773 SCSI_PROT_IP_CHECKSUM,
776 return flag_mask[prot_op];
779 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
780 unsigned int dix, unsigned int dif)
782 struct bio *bio = scmd->request->bio;
783 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
784 unsigned int protect = 0;
786 if (dix) { /* DIX Type 0, 1, 2, 3 */
787 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
788 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
790 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
791 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
794 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
795 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
797 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
798 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
801 if (dif) { /* DIX/DIF Type 1, 2, 3 */
802 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
804 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
805 protect = 3 << 5; /* Disable target PI checking */
806 else
807 protect = 1 << 5; /* Enable target PI checking */
810 scsi_set_prot_op(scmd, prot_op);
811 scsi_set_prot_type(scmd, dif);
812 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
814 return protect;
817 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
819 struct request_queue *q = sdkp->disk->queue;
820 unsigned int logical_block_size = sdkp->device->sector_size;
821 unsigned int max_blocks = 0;
823 q->limits.discard_alignment =
824 sdkp->unmap_alignment * logical_block_size;
825 q->limits.discard_granularity =
826 max(sdkp->physical_block_size,
827 sdkp->unmap_granularity * logical_block_size);
828 sdkp->provisioning_mode = mode;
830 switch (mode) {
832 case SD_LBP_FULL:
833 case SD_LBP_DISABLE:
834 blk_queue_max_discard_sectors(q, 0);
835 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
836 return;
838 case SD_LBP_UNMAP:
839 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
840 (u32)SD_MAX_WS16_BLOCKS);
841 break;
843 case SD_LBP_WS16:
844 if (sdkp->device->unmap_limit_for_ws)
845 max_blocks = sdkp->max_unmap_blocks;
846 else
847 max_blocks = sdkp->max_ws_blocks;
849 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
850 break;
852 case SD_LBP_WS10:
853 if (sdkp->device->unmap_limit_for_ws)
854 max_blocks = sdkp->max_unmap_blocks;
855 else
856 max_blocks = sdkp->max_ws_blocks;
858 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
859 break;
861 case SD_LBP_ZERO:
862 max_blocks = min_not_zero(sdkp->max_ws_blocks,
863 (u32)SD_MAX_WS10_BLOCKS);
864 break;
867 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
868 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
871 static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
873 struct scsi_device *sdp = cmd->device;
874 struct request *rq = cmd->request;
875 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
876 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
877 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
878 unsigned int data_len = 24;
879 char *buf;
881 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 if (!rq->special_vec.bv_page)
883 return BLK_STS_RESOURCE;
884 clear_highpage(rq->special_vec.bv_page);
885 rq->special_vec.bv_offset = 0;
886 rq->special_vec.bv_len = data_len;
887 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
889 cmd->cmd_len = 10;
890 cmd->cmnd[0] = UNMAP;
891 cmd->cmnd[8] = 24;
893 buf = page_address(rq->special_vec.bv_page);
894 put_unaligned_be16(6 + 16, &buf[0]);
895 put_unaligned_be16(16, &buf[2]);
896 put_unaligned_be64(lba, &buf[8]);
897 put_unaligned_be32(nr_blocks, &buf[16]);
899 cmd->allowed = sdkp->max_retries;
900 cmd->transfersize = data_len;
901 rq->timeout = SD_TIMEOUT;
903 return scsi_alloc_sgtables(cmd);
906 static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
907 bool unmap)
909 struct scsi_device *sdp = cmd->device;
910 struct request *rq = cmd->request;
911 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
912 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
913 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
914 u32 data_len = sdp->sector_size;
916 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
917 if (!rq->special_vec.bv_page)
918 return BLK_STS_RESOURCE;
919 clear_highpage(rq->special_vec.bv_page);
920 rq->special_vec.bv_offset = 0;
921 rq->special_vec.bv_len = data_len;
922 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
924 cmd->cmd_len = 16;
925 cmd->cmnd[0] = WRITE_SAME_16;
926 if (unmap)
927 cmd->cmnd[1] = 0x8; /* UNMAP */
928 put_unaligned_be64(lba, &cmd->cmnd[2]);
929 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
931 cmd->allowed = sdkp->max_retries;
932 cmd->transfersize = data_len;
933 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
935 return scsi_alloc_sgtables(cmd);
938 static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
939 bool unmap)
941 struct scsi_device *sdp = cmd->device;
942 struct request *rq = cmd->request;
943 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
944 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
945 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
946 u32 data_len = sdp->sector_size;
948 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
949 if (!rq->special_vec.bv_page)
950 return BLK_STS_RESOURCE;
951 clear_highpage(rq->special_vec.bv_page);
952 rq->special_vec.bv_offset = 0;
953 rq->special_vec.bv_len = data_len;
954 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
956 cmd->cmd_len = 10;
957 cmd->cmnd[0] = WRITE_SAME;
958 if (unmap)
959 cmd->cmnd[1] = 0x8; /* UNMAP */
960 put_unaligned_be32(lba, &cmd->cmnd[2]);
961 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
963 cmd->allowed = sdkp->max_retries;
964 cmd->transfersize = data_len;
965 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
967 return scsi_alloc_sgtables(cmd);
970 static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
972 struct request *rq = cmd->request;
973 struct scsi_device *sdp = cmd->device;
974 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
975 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
976 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
978 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
979 switch (sdkp->zeroing_mode) {
980 case SD_ZERO_WS16_UNMAP:
981 return sd_setup_write_same16_cmnd(cmd, true);
982 case SD_ZERO_WS10_UNMAP:
983 return sd_setup_write_same10_cmnd(cmd, true);
987 if (sdp->no_write_same)
988 return BLK_STS_TARGET;
990 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
991 return sd_setup_write_same16_cmnd(cmd, false);
993 return sd_setup_write_same10_cmnd(cmd, false);
996 static void sd_config_write_same(struct scsi_disk *sdkp)
998 struct request_queue *q = sdkp->disk->queue;
999 unsigned int logical_block_size = sdkp->device->sector_size;
1001 if (sdkp->device->no_write_same) {
1002 sdkp->max_ws_blocks = 0;
1003 goto out;
1006 /* Some devices can not handle block counts above 0xffff despite
1007 * supporting WRITE SAME(16). Consequently we default to 64k
1008 * blocks per I/O unless the device explicitly advertises a
1009 * bigger limit.
1011 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1012 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1013 (u32)SD_MAX_WS16_BLOCKS);
1014 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1015 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1016 (u32)SD_MAX_WS10_BLOCKS);
1017 else {
1018 sdkp->device->no_write_same = 1;
1019 sdkp->max_ws_blocks = 0;
1022 if (sdkp->lbprz && sdkp->lbpws)
1023 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1024 else if (sdkp->lbprz && sdkp->lbpws10)
1025 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1026 else if (sdkp->max_ws_blocks)
1027 sdkp->zeroing_mode = SD_ZERO_WS;
1028 else
1029 sdkp->zeroing_mode = SD_ZERO_WRITE;
1031 if (sdkp->max_ws_blocks &&
1032 sdkp->physical_block_size > logical_block_size) {
1034 * Reporting a maximum number of blocks that is not aligned
1035 * on the device physical size would cause a large write same
1036 * request to be split into physically unaligned chunks by
1037 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1038 * even if the caller of these functions took care to align the
1039 * large request. So make sure the maximum reported is aligned
1040 * to the device physical block size. This is only an optional
1041 * optimization for regular disks, but this is mandatory to
1042 * avoid failure of large write same requests directed at
1043 * sequential write required zones of host-managed ZBC disks.
1045 sdkp->max_ws_blocks =
1046 round_down(sdkp->max_ws_blocks,
1047 bytes_to_logical(sdkp->device,
1048 sdkp->physical_block_size));
1051 out:
1052 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1053 (logical_block_size >> 9));
1054 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1055 (logical_block_size >> 9));
1059 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1060 * @cmd: command to prepare
1062 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1063 * the preference indicated by the target device.
1065 static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1067 struct request *rq = cmd->request;
1068 struct scsi_device *sdp = cmd->device;
1069 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1070 struct bio *bio = rq->bio;
1071 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1072 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1073 blk_status_t ret;
1075 if (sdkp->device->no_write_same)
1076 return BLK_STS_TARGET;
1078 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1080 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1082 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1083 cmd->cmd_len = 16;
1084 cmd->cmnd[0] = WRITE_SAME_16;
1085 put_unaligned_be64(lba, &cmd->cmnd[2]);
1086 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1087 } else {
1088 cmd->cmd_len = 10;
1089 cmd->cmnd[0] = WRITE_SAME;
1090 put_unaligned_be32(lba, &cmd->cmnd[2]);
1091 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1094 cmd->transfersize = sdp->sector_size;
1095 cmd->allowed = sdkp->max_retries;
1098 * For WRITE SAME the data transferred via the DATA OUT buffer is
1099 * different from the amount of data actually written to the target.
1101 * We set up __data_len to the amount of data transferred via the
1102 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1103 * to transfer a single sector of data first, but then reset it to
1104 * the amount of data to be written right after so that the I/O path
1105 * knows how much to actually write.
1107 rq->__data_len = sdp->sector_size;
1108 ret = scsi_alloc_sgtables(cmd);
1109 rq->__data_len = blk_rq_bytes(rq);
1111 return ret;
1114 static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1116 struct request *rq = cmd->request;
1117 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1119 /* flush requests don't perform I/O, zero the S/G table */
1120 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1122 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1123 cmd->cmd_len = 10;
1124 cmd->transfersize = 0;
1125 cmd->allowed = sdkp->max_retries;
1127 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1128 return BLK_STS_OK;
1131 static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1132 sector_t lba, unsigned int nr_blocks,
1133 unsigned char flags)
1135 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1136 if (unlikely(cmd->cmnd == NULL))
1137 return BLK_STS_RESOURCE;
1139 cmd->cmd_len = SD_EXT_CDB_SIZE;
1140 memset(cmd->cmnd, 0, cmd->cmd_len);
1142 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1143 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1144 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1145 cmd->cmnd[10] = flags;
1146 put_unaligned_be64(lba, &cmd->cmnd[12]);
1147 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1148 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1150 return BLK_STS_OK;
1153 static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1154 sector_t lba, unsigned int nr_blocks,
1155 unsigned char flags)
1157 cmd->cmd_len = 16;
1158 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1159 cmd->cmnd[1] = flags;
1160 cmd->cmnd[14] = 0;
1161 cmd->cmnd[15] = 0;
1162 put_unaligned_be64(lba, &cmd->cmnd[2]);
1163 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1165 return BLK_STS_OK;
1168 static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1169 sector_t lba, unsigned int nr_blocks,
1170 unsigned char flags)
1172 cmd->cmd_len = 10;
1173 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1174 cmd->cmnd[1] = flags;
1175 cmd->cmnd[6] = 0;
1176 cmd->cmnd[9] = 0;
1177 put_unaligned_be32(lba, &cmd->cmnd[2]);
1178 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1180 return BLK_STS_OK;
1183 static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1184 sector_t lba, unsigned int nr_blocks,
1185 unsigned char flags)
1187 /* Avoid that 0 blocks gets translated into 256 blocks. */
1188 if (WARN_ON_ONCE(nr_blocks == 0))
1189 return BLK_STS_IOERR;
1191 if (unlikely(flags & 0x8)) {
1193 * This happens only if this drive failed 10byte rw
1194 * command with ILLEGAL_REQUEST during operation and
1195 * thus turned off use_10_for_rw.
1197 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1198 return BLK_STS_IOERR;
1201 cmd->cmd_len = 6;
1202 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1203 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1204 cmd->cmnd[2] = (lba >> 8) & 0xff;
1205 cmd->cmnd[3] = lba & 0xff;
1206 cmd->cmnd[4] = nr_blocks;
1207 cmd->cmnd[5] = 0;
1209 return BLK_STS_OK;
1212 static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1214 struct request *rq = cmd->request;
1215 struct scsi_device *sdp = cmd->device;
1216 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1217 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1218 sector_t threshold;
1219 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1220 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1221 bool write = rq_data_dir(rq) == WRITE;
1222 unsigned char protect, fua;
1223 blk_status_t ret;
1224 unsigned int dif;
1225 bool dix;
1227 ret = scsi_alloc_sgtables(cmd);
1228 if (ret != BLK_STS_OK)
1229 return ret;
1231 ret = BLK_STS_IOERR;
1232 if (!scsi_device_online(sdp) || sdp->changed) {
1233 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1234 goto fail;
1237 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1238 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1239 goto fail;
1242 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1243 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1244 goto fail;
1248 * Some SD card readers can't handle accesses which touch the
1249 * last one or two logical blocks. Split accesses as needed.
1251 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1253 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1254 if (lba < threshold) {
1255 /* Access up to the threshold but not beyond */
1256 nr_blocks = threshold - lba;
1257 } else {
1258 /* Access only a single logical block */
1259 nr_blocks = 1;
1263 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1264 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1265 if (ret)
1266 goto fail;
1269 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1270 dix = scsi_prot_sg_count(cmd);
1271 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1273 if (dif || dix)
1274 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1275 else
1276 protect = 0;
1278 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1279 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1280 protect | fua);
1281 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1282 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1283 protect | fua);
1284 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1285 sdp->use_10_for_rw || protect) {
1286 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1287 protect | fua);
1288 } else {
1289 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1290 protect | fua);
1293 if (unlikely(ret != BLK_STS_OK))
1294 goto fail;
1297 * We shouldn't disconnect in the middle of a sector, so with a dumb
1298 * host adapter, it's safe to assume that we can at least transfer
1299 * this many bytes between each connect / disconnect.
1301 cmd->transfersize = sdp->sector_size;
1302 cmd->underflow = nr_blocks << 9;
1303 cmd->allowed = sdkp->max_retries;
1304 cmd->sdb.length = nr_blocks * sdp->sector_size;
1306 SCSI_LOG_HLQUEUE(1,
1307 scmd_printk(KERN_INFO, cmd,
1308 "%s: block=%llu, count=%d\n", __func__,
1309 (unsigned long long)blk_rq_pos(rq),
1310 blk_rq_sectors(rq)));
1311 SCSI_LOG_HLQUEUE(2,
1312 scmd_printk(KERN_INFO, cmd,
1313 "%s %d/%u 512 byte blocks.\n",
1314 write ? "writing" : "reading", nr_blocks,
1315 blk_rq_sectors(rq)));
1318 * This indicates that the command is ready from our end to be queued.
1320 return BLK_STS_OK;
1321 fail:
1322 scsi_free_sgtables(cmd);
1323 return ret;
1326 static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1328 struct request *rq = cmd->request;
1330 switch (req_op(rq)) {
1331 case REQ_OP_DISCARD:
1332 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1333 case SD_LBP_UNMAP:
1334 return sd_setup_unmap_cmnd(cmd);
1335 case SD_LBP_WS16:
1336 return sd_setup_write_same16_cmnd(cmd, true);
1337 case SD_LBP_WS10:
1338 return sd_setup_write_same10_cmnd(cmd, true);
1339 case SD_LBP_ZERO:
1340 return sd_setup_write_same10_cmnd(cmd, false);
1341 default:
1342 return BLK_STS_TARGET;
1344 case REQ_OP_WRITE_ZEROES:
1345 return sd_setup_write_zeroes_cmnd(cmd);
1346 case REQ_OP_WRITE_SAME:
1347 return sd_setup_write_same_cmnd(cmd);
1348 case REQ_OP_FLUSH:
1349 return sd_setup_flush_cmnd(cmd);
1350 case REQ_OP_READ:
1351 case REQ_OP_WRITE:
1352 case REQ_OP_ZONE_APPEND:
1353 return sd_setup_read_write_cmnd(cmd);
1354 case REQ_OP_ZONE_RESET:
1355 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1356 false);
1357 case REQ_OP_ZONE_RESET_ALL:
1358 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1359 true);
1360 case REQ_OP_ZONE_OPEN:
1361 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1362 case REQ_OP_ZONE_CLOSE:
1363 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1364 case REQ_OP_ZONE_FINISH:
1365 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1366 default:
1367 WARN_ON_ONCE(1);
1368 return BLK_STS_NOTSUPP;
1372 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1374 struct request *rq = SCpnt->request;
1375 u8 *cmnd;
1377 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1378 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1380 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1381 cmnd = SCpnt->cmnd;
1382 SCpnt->cmnd = NULL;
1383 SCpnt->cmd_len = 0;
1384 mempool_free(cmnd, sd_cdb_pool);
1389 * sd_open - open a scsi disk device
1390 * @bdev: Block device of the scsi disk to open
1391 * @mode: FMODE_* mask
1393 * Returns 0 if successful. Returns a negated errno value in case
1394 * of error.
1396 * Note: This can be called from a user context (e.g. fsck(1) )
1397 * or from within the kernel (e.g. as a result of a mount(1) ).
1398 * In the latter case @inode and @filp carry an abridged amount
1399 * of information as noted above.
1401 * Locking: called with bdev->bd_mutex held.
1403 static int sd_open(struct block_device *bdev, fmode_t mode)
1405 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1406 struct scsi_device *sdev;
1407 int retval;
1409 if (!sdkp)
1410 return -ENXIO;
1412 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1414 sdev = sdkp->device;
1417 * If the device is in error recovery, wait until it is done.
1418 * If the device is offline, then disallow any access to it.
1420 retval = -ENXIO;
1421 if (!scsi_block_when_processing_errors(sdev))
1422 goto error_out;
1424 if (sdev->removable || sdkp->write_prot) {
1425 if (bdev_check_media_change(bdev))
1426 sd_revalidate_disk(bdev->bd_disk);
1430 * If the drive is empty, just let the open fail.
1432 retval = -ENOMEDIUM;
1433 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1434 goto error_out;
1437 * If the device has the write protect tab set, have the open fail
1438 * if the user expects to be able to write to the thing.
1440 retval = -EROFS;
1441 if (sdkp->write_prot && (mode & FMODE_WRITE))
1442 goto error_out;
1445 * It is possible that the disk changing stuff resulted in
1446 * the device being taken offline. If this is the case,
1447 * report this to the user, and don't pretend that the
1448 * open actually succeeded.
1450 retval = -ENXIO;
1451 if (!scsi_device_online(sdev))
1452 goto error_out;
1454 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1455 if (scsi_block_when_processing_errors(sdev))
1456 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1459 return 0;
1461 error_out:
1462 scsi_disk_put(sdkp);
1463 return retval;
1467 * sd_release - invoked when the (last) close(2) is called on this
1468 * scsi disk.
1469 * @disk: disk to release
1470 * @mode: FMODE_* mask
1472 * Returns 0.
1474 * Note: may block (uninterruptible) if error recovery is underway
1475 * on this disk.
1477 * Locking: called with bdev->bd_mutex held.
1479 static void sd_release(struct gendisk *disk, fmode_t mode)
1481 struct scsi_disk *sdkp = scsi_disk(disk);
1482 struct scsi_device *sdev = sdkp->device;
1484 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1486 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1487 if (scsi_block_when_processing_errors(sdev))
1488 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1491 scsi_disk_put(sdkp);
1494 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1496 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1497 struct scsi_device *sdp = sdkp->device;
1498 struct Scsi_Host *host = sdp->host;
1499 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1500 int diskinfo[4];
1502 /* default to most commonly used values */
1503 diskinfo[0] = 0x40; /* 1 << 6 */
1504 diskinfo[1] = 0x20; /* 1 << 5 */
1505 diskinfo[2] = capacity >> 11;
1507 /* override with calculated, extended default, or driver values */
1508 if (host->hostt->bios_param)
1509 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1510 else
1511 scsicam_bios_param(bdev, capacity, diskinfo);
1513 geo->heads = diskinfo[0];
1514 geo->sectors = diskinfo[1];
1515 geo->cylinders = diskinfo[2];
1516 return 0;
1520 * sd_ioctl - process an ioctl
1521 * @bdev: target block device
1522 * @mode: FMODE_* mask
1523 * @cmd: ioctl command number
1524 * @p: this is third argument given to ioctl(2) system call.
1525 * Often contains a pointer.
1527 * Returns 0 if successful (some ioctls return positive numbers on
1528 * success as well). Returns a negated errno value in case of error.
1530 * Note: most ioctls are forward onto the block subsystem or further
1531 * down in the scsi subsystem.
1533 static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1534 unsigned int cmd, void __user *p)
1536 struct gendisk *disk = bdev->bd_disk;
1537 struct scsi_disk *sdkp = scsi_disk(disk);
1538 struct scsi_device *sdp = sdkp->device;
1539 int error;
1541 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1542 "cmd=0x%x\n", disk->disk_name, cmd));
1544 error = scsi_verify_blk_ioctl(bdev, cmd);
1545 if (error < 0)
1546 return error;
1549 * If we are in the middle of error recovery, don't let anyone
1550 * else try and use this device. Also, if error recovery fails, it
1551 * may try and take the device offline, in which case all further
1552 * access to the device is prohibited.
1554 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1555 (mode & FMODE_NDELAY) != 0);
1556 if (error)
1557 goto out;
1559 if (is_sed_ioctl(cmd))
1560 return sed_ioctl(sdkp->opal_dev, cmd, p);
1563 * Send SCSI addressing ioctls directly to mid level, send other
1564 * ioctls to block level and then onto mid level if they can't be
1565 * resolved.
1567 switch (cmd) {
1568 case SCSI_IOCTL_GET_IDLUN:
1569 case SCSI_IOCTL_GET_BUS_NUMBER:
1570 error = scsi_ioctl(sdp, cmd, p);
1571 break;
1572 default:
1573 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1574 break;
1576 out:
1577 return error;
1580 static void set_media_not_present(struct scsi_disk *sdkp)
1582 if (sdkp->media_present)
1583 sdkp->device->changed = 1;
1585 if (sdkp->device->removable) {
1586 sdkp->media_present = 0;
1587 sdkp->capacity = 0;
1591 static int media_not_present(struct scsi_disk *sdkp,
1592 struct scsi_sense_hdr *sshdr)
1594 if (!scsi_sense_valid(sshdr))
1595 return 0;
1597 /* not invoked for commands that could return deferred errors */
1598 switch (sshdr->sense_key) {
1599 case UNIT_ATTENTION:
1600 case NOT_READY:
1601 /* medium not present */
1602 if (sshdr->asc == 0x3A) {
1603 set_media_not_present(sdkp);
1604 return 1;
1607 return 0;
1611 * sd_check_events - check media events
1612 * @disk: kernel device descriptor
1613 * @clearing: disk events currently being cleared
1615 * Returns mask of DISK_EVENT_*.
1617 * Note: this function is invoked from the block subsystem.
1619 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1621 struct scsi_disk *sdkp = scsi_disk_get(disk);
1622 struct scsi_device *sdp;
1623 int retval;
1625 if (!sdkp)
1626 return 0;
1628 sdp = sdkp->device;
1629 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1632 * If the device is offline, don't send any commands - just pretend as
1633 * if the command failed. If the device ever comes back online, we
1634 * can deal with it then. It is only because of unrecoverable errors
1635 * that we would ever take a device offline in the first place.
1637 if (!scsi_device_online(sdp)) {
1638 set_media_not_present(sdkp);
1639 goto out;
1643 * Using TEST_UNIT_READY enables differentiation between drive with
1644 * no cartridge loaded - NOT READY, drive with changed cartridge -
1645 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1647 * Drives that auto spin down. eg iomega jaz 1G, will be started
1648 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1649 * sd_revalidate() is called.
1651 if (scsi_block_when_processing_errors(sdp)) {
1652 struct scsi_sense_hdr sshdr = { 0, };
1654 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1655 &sshdr);
1657 /* failed to execute TUR, assume media not present */
1658 if (host_byte(retval)) {
1659 set_media_not_present(sdkp);
1660 goto out;
1663 if (media_not_present(sdkp, &sshdr))
1664 goto out;
1668 * For removable scsi disk we have to recognise the presence
1669 * of a disk in the drive.
1671 if (!sdkp->media_present)
1672 sdp->changed = 1;
1673 sdkp->media_present = 1;
1674 out:
1676 * sdp->changed is set under the following conditions:
1678 * Medium present state has changed in either direction.
1679 * Device has indicated UNIT_ATTENTION.
1681 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1682 sdp->changed = 0;
1683 scsi_disk_put(sdkp);
1684 return retval;
1687 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1689 int retries, res;
1690 struct scsi_device *sdp = sdkp->device;
1691 const int timeout = sdp->request_queue->rq_timeout
1692 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1693 struct scsi_sense_hdr my_sshdr;
1695 if (!scsi_device_online(sdp))
1696 return -ENODEV;
1698 /* caller might not be interested in sense, but we need it */
1699 if (!sshdr)
1700 sshdr = &my_sshdr;
1702 for (retries = 3; retries > 0; --retries) {
1703 unsigned char cmd[10] = { 0 };
1705 cmd[0] = SYNCHRONIZE_CACHE;
1707 * Leave the rest of the command zero to indicate
1708 * flush everything.
1710 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1711 timeout, sdkp->max_retries, 0, RQF_PM, NULL);
1712 if (res == 0)
1713 break;
1716 if (res) {
1717 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1719 if (driver_byte(res) == DRIVER_SENSE)
1720 sd_print_sense_hdr(sdkp, sshdr);
1722 /* we need to evaluate the error return */
1723 if (scsi_sense_valid(sshdr) &&
1724 (sshdr->asc == 0x3a || /* medium not present */
1725 sshdr->asc == 0x20 || /* invalid command */
1726 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1727 /* this is no error here */
1728 return 0;
1730 switch (host_byte(res)) {
1731 /* ignore errors due to racing a disconnection */
1732 case DID_BAD_TARGET:
1733 case DID_NO_CONNECT:
1734 return 0;
1735 /* signal the upper layer it might try again */
1736 case DID_BUS_BUSY:
1737 case DID_IMM_RETRY:
1738 case DID_REQUEUE:
1739 case DID_SOFT_ERROR:
1740 return -EBUSY;
1741 default:
1742 return -EIO;
1745 return 0;
1748 static void sd_rescan(struct device *dev)
1750 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1752 sd_revalidate_disk(sdkp->disk);
1755 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1756 unsigned int cmd, unsigned long arg)
1758 void __user *p = (void __user *)arg;
1759 int ret;
1761 ret = sd_ioctl_common(bdev, mode, cmd, p);
1762 if (ret != -ENOTTY)
1763 return ret;
1765 return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1768 #ifdef CONFIG_COMPAT
1769 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1770 unsigned int cmd, unsigned long arg)
1772 void __user *p = compat_ptr(arg);
1773 int ret;
1775 ret = sd_ioctl_common(bdev, mode, cmd, p);
1776 if (ret != -ENOTTY)
1777 return ret;
1779 return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1781 #endif
1783 static char sd_pr_type(enum pr_type type)
1785 switch (type) {
1786 case PR_WRITE_EXCLUSIVE:
1787 return 0x01;
1788 case PR_EXCLUSIVE_ACCESS:
1789 return 0x03;
1790 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1791 return 0x05;
1792 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1793 return 0x06;
1794 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1795 return 0x07;
1796 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1797 return 0x08;
1798 default:
1799 return 0;
1803 static int sd_pr_command(struct block_device *bdev, u8 sa,
1804 u64 key, u64 sa_key, u8 type, u8 flags)
1806 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1807 struct scsi_device *sdev = sdkp->device;
1808 struct scsi_sense_hdr sshdr;
1809 int result;
1810 u8 cmd[16] = { 0, };
1811 u8 data[24] = { 0, };
1813 cmd[0] = PERSISTENT_RESERVE_OUT;
1814 cmd[1] = sa;
1815 cmd[2] = type;
1816 put_unaligned_be32(sizeof(data), &cmd[5]);
1818 put_unaligned_be64(key, &data[0]);
1819 put_unaligned_be64(sa_key, &data[8]);
1820 data[20] = flags;
1822 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1823 &sshdr, SD_TIMEOUT, sdkp->max_retries, NULL);
1825 if (driver_byte(result) == DRIVER_SENSE &&
1826 scsi_sense_valid(&sshdr)) {
1827 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1828 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1831 return result;
1834 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1835 u32 flags)
1837 if (flags & ~PR_FL_IGNORE_KEY)
1838 return -EOPNOTSUPP;
1839 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1840 old_key, new_key, 0,
1841 (1 << 0) /* APTPL */);
1844 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1845 u32 flags)
1847 if (flags)
1848 return -EOPNOTSUPP;
1849 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1852 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1854 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1857 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1858 enum pr_type type, bool abort)
1860 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1861 sd_pr_type(type), 0);
1864 static int sd_pr_clear(struct block_device *bdev, u64 key)
1866 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1869 static const struct pr_ops sd_pr_ops = {
1870 .pr_register = sd_pr_register,
1871 .pr_reserve = sd_pr_reserve,
1872 .pr_release = sd_pr_release,
1873 .pr_preempt = sd_pr_preempt,
1874 .pr_clear = sd_pr_clear,
1877 static const struct block_device_operations sd_fops = {
1878 .owner = THIS_MODULE,
1879 .open = sd_open,
1880 .release = sd_release,
1881 .ioctl = sd_ioctl,
1882 .getgeo = sd_getgeo,
1883 #ifdef CONFIG_COMPAT
1884 .compat_ioctl = sd_compat_ioctl,
1885 #endif
1886 .check_events = sd_check_events,
1887 .unlock_native_capacity = sd_unlock_native_capacity,
1888 .report_zones = sd_zbc_report_zones,
1889 .pr_ops = &sd_pr_ops,
1893 * sd_eh_reset - reset error handling callback
1894 * @scmd: sd-issued command that has failed
1896 * This function is called by the SCSI midlayer before starting
1897 * SCSI EH. When counting medium access failures we have to be
1898 * careful to register it only only once per device and SCSI EH run;
1899 * there might be several timed out commands which will cause the
1900 * 'max_medium_access_timeouts' counter to trigger after the first
1901 * SCSI EH run already and set the device to offline.
1902 * So this function resets the internal counter before starting SCSI EH.
1904 static void sd_eh_reset(struct scsi_cmnd *scmd)
1906 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1908 /* New SCSI EH run, reset gate variable */
1909 sdkp->ignore_medium_access_errors = false;
1913 * sd_eh_action - error handling callback
1914 * @scmd: sd-issued command that has failed
1915 * @eh_disp: The recovery disposition suggested by the midlayer
1917 * This function is called by the SCSI midlayer upon completion of an
1918 * error test command (currently TEST UNIT READY). The result of sending
1919 * the eh command is passed in eh_disp. We're looking for devices that
1920 * fail medium access commands but are OK with non access commands like
1921 * test unit ready (so wrongly see the device as having a successful
1922 * recovery)
1924 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1926 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1927 struct scsi_device *sdev = scmd->device;
1929 if (!scsi_device_online(sdev) ||
1930 !scsi_medium_access_command(scmd) ||
1931 host_byte(scmd->result) != DID_TIME_OUT ||
1932 eh_disp != SUCCESS)
1933 return eh_disp;
1936 * The device has timed out executing a medium access command.
1937 * However, the TEST UNIT READY command sent during error
1938 * handling completed successfully. Either the device is in the
1939 * process of recovering or has it suffered an internal failure
1940 * that prevents access to the storage medium.
1942 if (!sdkp->ignore_medium_access_errors) {
1943 sdkp->medium_access_timed_out++;
1944 sdkp->ignore_medium_access_errors = true;
1948 * If the device keeps failing read/write commands but TEST UNIT
1949 * READY always completes successfully we assume that medium
1950 * access is no longer possible and take the device offline.
1952 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1953 scmd_printk(KERN_ERR, scmd,
1954 "Medium access timeout failure. Offlining disk!\n");
1955 mutex_lock(&sdev->state_mutex);
1956 scsi_device_set_state(sdev, SDEV_OFFLINE);
1957 mutex_unlock(&sdev->state_mutex);
1959 return SUCCESS;
1962 return eh_disp;
1965 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1967 struct request *req = scmd->request;
1968 struct scsi_device *sdev = scmd->device;
1969 unsigned int transferred, good_bytes;
1970 u64 start_lba, end_lba, bad_lba;
1973 * Some commands have a payload smaller than the device logical
1974 * block size (e.g. INQUIRY on a 4K disk).
1976 if (scsi_bufflen(scmd) <= sdev->sector_size)
1977 return 0;
1979 /* Check if we have a 'bad_lba' information */
1980 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1981 SCSI_SENSE_BUFFERSIZE,
1982 &bad_lba))
1983 return 0;
1986 * If the bad lba was reported incorrectly, we have no idea where
1987 * the error is.
1989 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1990 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1991 if (bad_lba < start_lba || bad_lba >= end_lba)
1992 return 0;
1995 * resid is optional but mostly filled in. When it's unused,
1996 * its value is zero, so we assume the whole buffer transferred
1998 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2000 /* This computation should always be done in terms of the
2001 * resolution of the device's medium.
2003 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2005 return min(good_bytes, transferred);
2009 * sd_done - bottom half handler: called when the lower level
2010 * driver has completed (successfully or otherwise) a scsi command.
2011 * @SCpnt: mid-level's per command structure.
2013 * Note: potentially run from within an ISR. Must not block.
2015 static int sd_done(struct scsi_cmnd *SCpnt)
2017 int result = SCpnt->result;
2018 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2019 unsigned int sector_size = SCpnt->device->sector_size;
2020 unsigned int resid;
2021 struct scsi_sense_hdr sshdr;
2022 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
2023 struct request *req = SCpnt->request;
2024 int sense_valid = 0;
2025 int sense_deferred = 0;
2027 switch (req_op(req)) {
2028 case REQ_OP_DISCARD:
2029 case REQ_OP_WRITE_ZEROES:
2030 case REQ_OP_WRITE_SAME:
2031 case REQ_OP_ZONE_RESET:
2032 case REQ_OP_ZONE_RESET_ALL:
2033 case REQ_OP_ZONE_OPEN:
2034 case REQ_OP_ZONE_CLOSE:
2035 case REQ_OP_ZONE_FINISH:
2036 if (!result) {
2037 good_bytes = blk_rq_bytes(req);
2038 scsi_set_resid(SCpnt, 0);
2039 } else {
2040 good_bytes = 0;
2041 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2043 break;
2044 default:
2046 * In case of bogus fw or device, we could end up having
2047 * an unaligned partial completion. Check this here and force
2048 * alignment.
2050 resid = scsi_get_resid(SCpnt);
2051 if (resid & (sector_size - 1)) {
2052 sd_printk(KERN_INFO, sdkp,
2053 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2054 resid, sector_size);
2055 scsi_print_command(SCpnt);
2056 resid = min(scsi_bufflen(SCpnt),
2057 round_up(resid, sector_size));
2058 scsi_set_resid(SCpnt, resid);
2062 if (result) {
2063 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2064 if (sense_valid)
2065 sense_deferred = scsi_sense_is_deferred(&sshdr);
2067 sdkp->medium_access_timed_out = 0;
2069 if (driver_byte(result) != DRIVER_SENSE &&
2070 (!sense_valid || sense_deferred))
2071 goto out;
2073 switch (sshdr.sense_key) {
2074 case HARDWARE_ERROR:
2075 case MEDIUM_ERROR:
2076 good_bytes = sd_completed_bytes(SCpnt);
2077 break;
2078 case RECOVERED_ERROR:
2079 good_bytes = scsi_bufflen(SCpnt);
2080 break;
2081 case NO_SENSE:
2082 /* This indicates a false check condition, so ignore it. An
2083 * unknown amount of data was transferred so treat it as an
2084 * error.
2086 SCpnt->result = 0;
2087 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2088 break;
2089 case ABORTED_COMMAND:
2090 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2091 good_bytes = sd_completed_bytes(SCpnt);
2092 break;
2093 case ILLEGAL_REQUEST:
2094 switch (sshdr.asc) {
2095 case 0x10: /* DIX: Host detected corruption */
2096 good_bytes = sd_completed_bytes(SCpnt);
2097 break;
2098 case 0x20: /* INVALID COMMAND OPCODE */
2099 case 0x24: /* INVALID FIELD IN CDB */
2100 switch (SCpnt->cmnd[0]) {
2101 case UNMAP:
2102 sd_config_discard(sdkp, SD_LBP_DISABLE);
2103 break;
2104 case WRITE_SAME_16:
2105 case WRITE_SAME:
2106 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2107 sd_config_discard(sdkp, SD_LBP_DISABLE);
2108 } else {
2109 sdkp->device->no_write_same = 1;
2110 sd_config_write_same(sdkp);
2111 req->rq_flags |= RQF_QUIET;
2113 break;
2116 break;
2117 default:
2118 break;
2121 out:
2122 if (sd_is_zoned(sdkp))
2123 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2125 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2126 "sd_done: completed %d of %d bytes\n",
2127 good_bytes, scsi_bufflen(SCpnt)));
2129 return good_bytes;
2133 * spinup disk - called only in sd_revalidate_disk()
2135 static void
2136 sd_spinup_disk(struct scsi_disk *sdkp)
2138 unsigned char cmd[10];
2139 unsigned long spintime_expire = 0;
2140 int retries, spintime;
2141 unsigned int the_result;
2142 struct scsi_sense_hdr sshdr;
2143 int sense_valid = 0;
2145 spintime = 0;
2147 /* Spin up drives, as required. Only do this at boot time */
2148 /* Spinup needs to be done for module loads too. */
2149 do {
2150 retries = 0;
2152 do {
2153 cmd[0] = TEST_UNIT_READY;
2154 memset((void *) &cmd[1], 0, 9);
2156 the_result = scsi_execute_req(sdkp->device, cmd,
2157 DMA_NONE, NULL, 0,
2158 &sshdr, SD_TIMEOUT,
2159 sdkp->max_retries, NULL);
2162 * If the drive has indicated to us that it
2163 * doesn't have any media in it, don't bother
2164 * with any more polling.
2166 if (media_not_present(sdkp, &sshdr))
2167 return;
2169 if (the_result)
2170 sense_valid = scsi_sense_valid(&sshdr);
2171 retries++;
2172 } while (retries < 3 &&
2173 (!scsi_status_is_good(the_result) ||
2174 ((driver_byte(the_result) == DRIVER_SENSE) &&
2175 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2177 if (driver_byte(the_result) != DRIVER_SENSE) {
2178 /* no sense, TUR either succeeded or failed
2179 * with a status error */
2180 if(!spintime && !scsi_status_is_good(the_result)) {
2181 sd_print_result(sdkp, "Test Unit Ready failed",
2182 the_result);
2184 break;
2188 * The device does not want the automatic start to be issued.
2190 if (sdkp->device->no_start_on_add)
2191 break;
2193 if (sense_valid && sshdr.sense_key == NOT_READY) {
2194 if (sshdr.asc == 4 && sshdr.ascq == 3)
2195 break; /* manual intervention required */
2196 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2197 break; /* standby */
2198 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2199 break; /* unavailable */
2200 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2201 break; /* sanitize in progress */
2203 * Issue command to spin up drive when not ready
2205 if (!spintime) {
2206 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2207 cmd[0] = START_STOP;
2208 cmd[1] = 1; /* Return immediately */
2209 memset((void *) &cmd[2], 0, 8);
2210 cmd[4] = 1; /* Start spin cycle */
2211 if (sdkp->device->start_stop_pwr_cond)
2212 cmd[4] |= 1 << 4;
2213 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2214 NULL, 0, &sshdr,
2215 SD_TIMEOUT, sdkp->max_retries,
2216 NULL);
2217 spintime_expire = jiffies + 100 * HZ;
2218 spintime = 1;
2220 /* Wait 1 second for next try */
2221 msleep(1000);
2222 printk(KERN_CONT ".");
2225 * Wait for USB flash devices with slow firmware.
2226 * Yes, this sense key/ASC combination shouldn't
2227 * occur here. It's characteristic of these devices.
2229 } else if (sense_valid &&
2230 sshdr.sense_key == UNIT_ATTENTION &&
2231 sshdr.asc == 0x28) {
2232 if (!spintime) {
2233 spintime_expire = jiffies + 5 * HZ;
2234 spintime = 1;
2236 /* Wait 1 second for next try */
2237 msleep(1000);
2238 } else {
2239 /* we don't understand the sense code, so it's
2240 * probably pointless to loop */
2241 if(!spintime) {
2242 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2243 sd_print_sense_hdr(sdkp, &sshdr);
2245 break;
2248 } while (spintime && time_before_eq(jiffies, spintime_expire));
2250 if (spintime) {
2251 if (scsi_status_is_good(the_result))
2252 printk(KERN_CONT "ready\n");
2253 else
2254 printk(KERN_CONT "not responding...\n");
2259 * Determine whether disk supports Data Integrity Field.
2261 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2263 struct scsi_device *sdp = sdkp->device;
2264 u8 type;
2265 int ret = 0;
2267 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2268 sdkp->protection_type = 0;
2269 return ret;
2272 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2274 if (type > T10_PI_TYPE3_PROTECTION)
2275 ret = -ENODEV;
2276 else if (scsi_host_dif_capable(sdp->host, type))
2277 ret = 1;
2279 if (sdkp->first_scan || type != sdkp->protection_type)
2280 switch (ret) {
2281 case -ENODEV:
2282 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2283 " protection type %u. Disabling disk!\n",
2284 type);
2285 break;
2286 case 1:
2287 sd_printk(KERN_NOTICE, sdkp,
2288 "Enabling DIF Type %u protection\n", type);
2289 break;
2290 case 0:
2291 sd_printk(KERN_NOTICE, sdkp,
2292 "Disabling DIF Type %u protection\n", type);
2293 break;
2296 sdkp->protection_type = type;
2298 return ret;
2301 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2302 struct scsi_sense_hdr *sshdr, int sense_valid,
2303 int the_result)
2305 if (driver_byte(the_result) == DRIVER_SENSE)
2306 sd_print_sense_hdr(sdkp, sshdr);
2307 else
2308 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2311 * Set dirty bit for removable devices if not ready -
2312 * sometimes drives will not report this properly.
2314 if (sdp->removable &&
2315 sense_valid && sshdr->sense_key == NOT_READY)
2316 set_media_not_present(sdkp);
2319 * We used to set media_present to 0 here to indicate no media
2320 * in the drive, but some drives fail read capacity even with
2321 * media present, so we can't do that.
2323 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2326 #define RC16_LEN 32
2327 #if RC16_LEN > SD_BUF_SIZE
2328 #error RC16_LEN must not be more than SD_BUF_SIZE
2329 #endif
2331 #define READ_CAPACITY_RETRIES_ON_RESET 10
2333 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2334 unsigned char *buffer)
2336 unsigned char cmd[16];
2337 struct scsi_sense_hdr sshdr;
2338 int sense_valid = 0;
2339 int the_result;
2340 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2341 unsigned int alignment;
2342 unsigned long long lba;
2343 unsigned sector_size;
2345 if (sdp->no_read_capacity_16)
2346 return -EINVAL;
2348 do {
2349 memset(cmd, 0, 16);
2350 cmd[0] = SERVICE_ACTION_IN_16;
2351 cmd[1] = SAI_READ_CAPACITY_16;
2352 cmd[13] = RC16_LEN;
2353 memset(buffer, 0, RC16_LEN);
2355 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2356 buffer, RC16_LEN, &sshdr,
2357 SD_TIMEOUT, sdkp->max_retries, NULL);
2359 if (media_not_present(sdkp, &sshdr))
2360 return -ENODEV;
2362 if (the_result) {
2363 sense_valid = scsi_sense_valid(&sshdr);
2364 if (sense_valid &&
2365 sshdr.sense_key == ILLEGAL_REQUEST &&
2366 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2367 sshdr.ascq == 0x00)
2368 /* Invalid Command Operation Code or
2369 * Invalid Field in CDB, just retry
2370 * silently with RC10 */
2371 return -EINVAL;
2372 if (sense_valid &&
2373 sshdr.sense_key == UNIT_ATTENTION &&
2374 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2375 /* Device reset might occur several times,
2376 * give it one more chance */
2377 if (--reset_retries > 0)
2378 continue;
2380 retries--;
2382 } while (the_result && retries);
2384 if (the_result) {
2385 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2386 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2387 return -EINVAL;
2390 sector_size = get_unaligned_be32(&buffer[8]);
2391 lba = get_unaligned_be64(&buffer[0]);
2393 if (sd_read_protection_type(sdkp, buffer) < 0) {
2394 sdkp->capacity = 0;
2395 return -ENODEV;
2398 /* Logical blocks per physical block exponent */
2399 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2401 /* RC basis */
2402 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2404 /* Lowest aligned logical block */
2405 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2406 blk_queue_alignment_offset(sdp->request_queue, alignment);
2407 if (alignment && sdkp->first_scan)
2408 sd_printk(KERN_NOTICE, sdkp,
2409 "physical block alignment offset: %u\n", alignment);
2411 if (buffer[14] & 0x80) { /* LBPME */
2412 sdkp->lbpme = 1;
2414 if (buffer[14] & 0x40) /* LBPRZ */
2415 sdkp->lbprz = 1;
2417 sd_config_discard(sdkp, SD_LBP_WS16);
2420 sdkp->capacity = lba + 1;
2421 return sector_size;
2424 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2425 unsigned char *buffer)
2427 unsigned char cmd[16];
2428 struct scsi_sense_hdr sshdr;
2429 int sense_valid = 0;
2430 int the_result;
2431 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2432 sector_t lba;
2433 unsigned sector_size;
2435 do {
2436 cmd[0] = READ_CAPACITY;
2437 memset(&cmd[1], 0, 9);
2438 memset(buffer, 0, 8);
2440 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2441 buffer, 8, &sshdr,
2442 SD_TIMEOUT, sdkp->max_retries, NULL);
2444 if (media_not_present(sdkp, &sshdr))
2445 return -ENODEV;
2447 if (the_result) {
2448 sense_valid = scsi_sense_valid(&sshdr);
2449 if (sense_valid &&
2450 sshdr.sense_key == UNIT_ATTENTION &&
2451 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2452 /* Device reset might occur several times,
2453 * give it one more chance */
2454 if (--reset_retries > 0)
2455 continue;
2457 retries--;
2459 } while (the_result && retries);
2461 if (the_result) {
2462 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2463 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2464 return -EINVAL;
2467 sector_size = get_unaligned_be32(&buffer[4]);
2468 lba = get_unaligned_be32(&buffer[0]);
2470 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2471 /* Some buggy (usb cardreader) devices return an lba of
2472 0xffffffff when the want to report a size of 0 (with
2473 which they really mean no media is present) */
2474 sdkp->capacity = 0;
2475 sdkp->physical_block_size = sector_size;
2476 return sector_size;
2479 sdkp->capacity = lba + 1;
2480 sdkp->physical_block_size = sector_size;
2481 return sector_size;
2484 static int sd_try_rc16_first(struct scsi_device *sdp)
2486 if (sdp->host->max_cmd_len < 16)
2487 return 0;
2488 if (sdp->try_rc_10_first)
2489 return 0;
2490 if (sdp->scsi_level > SCSI_SPC_2)
2491 return 1;
2492 if (scsi_device_protection(sdp))
2493 return 1;
2494 return 0;
2498 * read disk capacity
2500 static void
2501 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2503 int sector_size;
2504 struct scsi_device *sdp = sdkp->device;
2506 if (sd_try_rc16_first(sdp)) {
2507 sector_size = read_capacity_16(sdkp, sdp, buffer);
2508 if (sector_size == -EOVERFLOW)
2509 goto got_data;
2510 if (sector_size == -ENODEV)
2511 return;
2512 if (sector_size < 0)
2513 sector_size = read_capacity_10(sdkp, sdp, buffer);
2514 if (sector_size < 0)
2515 return;
2516 } else {
2517 sector_size = read_capacity_10(sdkp, sdp, buffer);
2518 if (sector_size == -EOVERFLOW)
2519 goto got_data;
2520 if (sector_size < 0)
2521 return;
2522 if ((sizeof(sdkp->capacity) > 4) &&
2523 (sdkp->capacity > 0xffffffffULL)) {
2524 int old_sector_size = sector_size;
2525 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2526 "Trying to use READ CAPACITY(16).\n");
2527 sector_size = read_capacity_16(sdkp, sdp, buffer);
2528 if (sector_size < 0) {
2529 sd_printk(KERN_NOTICE, sdkp,
2530 "Using 0xffffffff as device size\n");
2531 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2532 sector_size = old_sector_size;
2533 goto got_data;
2535 /* Remember that READ CAPACITY(16) succeeded */
2536 sdp->try_rc_10_first = 0;
2540 /* Some devices are known to return the total number of blocks,
2541 * not the highest block number. Some devices have versions
2542 * which do this and others which do not. Some devices we might
2543 * suspect of doing this but we don't know for certain.
2545 * If we know the reported capacity is wrong, decrement it. If
2546 * we can only guess, then assume the number of blocks is even
2547 * (usually true but not always) and err on the side of lowering
2548 * the capacity.
2550 if (sdp->fix_capacity ||
2551 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2552 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2553 "from its reported value: %llu\n",
2554 (unsigned long long) sdkp->capacity);
2555 --sdkp->capacity;
2558 got_data:
2559 if (sector_size == 0) {
2560 sector_size = 512;
2561 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2562 "assuming 512.\n");
2565 if (sector_size != 512 &&
2566 sector_size != 1024 &&
2567 sector_size != 2048 &&
2568 sector_size != 4096) {
2569 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2570 sector_size);
2572 * The user might want to re-format the drive with
2573 * a supported sectorsize. Once this happens, it
2574 * would be relatively trivial to set the thing up.
2575 * For this reason, we leave the thing in the table.
2577 sdkp->capacity = 0;
2579 * set a bogus sector size so the normal read/write
2580 * logic in the block layer will eventually refuse any
2581 * request on this device without tripping over power
2582 * of two sector size assumptions
2584 sector_size = 512;
2586 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2587 blk_queue_physical_block_size(sdp->request_queue,
2588 sdkp->physical_block_size);
2589 sdkp->device->sector_size = sector_size;
2591 if (sdkp->capacity > 0xffffffff)
2592 sdp->use_16_for_rw = 1;
2597 * Print disk capacity
2599 static void
2600 sd_print_capacity(struct scsi_disk *sdkp,
2601 sector_t old_capacity)
2603 int sector_size = sdkp->device->sector_size;
2604 char cap_str_2[10], cap_str_10[10];
2606 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2607 return;
2609 string_get_size(sdkp->capacity, sector_size,
2610 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2611 string_get_size(sdkp->capacity, sector_size,
2612 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2614 sd_printk(KERN_NOTICE, sdkp,
2615 "%llu %d-byte logical blocks: (%s/%s)\n",
2616 (unsigned long long)sdkp->capacity,
2617 sector_size, cap_str_10, cap_str_2);
2619 if (sdkp->physical_block_size != sector_size)
2620 sd_printk(KERN_NOTICE, sdkp,
2621 "%u-byte physical blocks\n",
2622 sdkp->physical_block_size);
2625 /* called with buffer of length 512 */
2626 static inline int
2627 sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2628 unsigned char *buffer, int len, struct scsi_mode_data *data,
2629 struct scsi_sense_hdr *sshdr)
2631 return scsi_mode_sense(sdkp->device, dbd, modepage, buffer, len,
2632 SD_TIMEOUT, sdkp->max_retries, data,
2633 sshdr);
2637 * read write protect setting, if possible - called only in sd_revalidate_disk()
2638 * called with buffer of length SD_BUF_SIZE
2640 static void
2641 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2643 int res;
2644 struct scsi_device *sdp = sdkp->device;
2645 struct scsi_mode_data data;
2646 int old_wp = sdkp->write_prot;
2648 set_disk_ro(sdkp->disk, 0);
2649 if (sdp->skip_ms_page_3f) {
2650 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2651 return;
2654 if (sdp->use_192_bytes_for_3f) {
2655 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2656 } else {
2658 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2659 * We have to start carefully: some devices hang if we ask
2660 * for more than is available.
2662 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2665 * Second attempt: ask for page 0 When only page 0 is
2666 * implemented, a request for page 3F may return Sense Key
2667 * 5: Illegal Request, Sense Code 24: Invalid field in
2668 * CDB.
2670 if (!scsi_status_is_good(res))
2671 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2674 * Third attempt: ask 255 bytes, as we did earlier.
2676 if (!scsi_status_is_good(res))
2677 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2678 &data, NULL);
2681 if (!scsi_status_is_good(res)) {
2682 sd_first_printk(KERN_WARNING, sdkp,
2683 "Test WP failed, assume Write Enabled\n");
2684 } else {
2685 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2686 set_disk_ro(sdkp->disk, sdkp->write_prot);
2687 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2688 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2689 sdkp->write_prot ? "on" : "off");
2690 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2696 * sd_read_cache_type - called only from sd_revalidate_disk()
2697 * called with buffer of length SD_BUF_SIZE
2699 static void
2700 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2702 int len = 0, res;
2703 struct scsi_device *sdp = sdkp->device;
2705 int dbd;
2706 int modepage;
2707 int first_len;
2708 struct scsi_mode_data data;
2709 struct scsi_sense_hdr sshdr;
2710 int old_wce = sdkp->WCE;
2711 int old_rcd = sdkp->RCD;
2712 int old_dpofua = sdkp->DPOFUA;
2715 if (sdkp->cache_override)
2716 return;
2718 first_len = 4;
2719 if (sdp->skip_ms_page_8) {
2720 if (sdp->type == TYPE_RBC)
2721 goto defaults;
2722 else {
2723 if (sdp->skip_ms_page_3f)
2724 goto defaults;
2725 modepage = 0x3F;
2726 if (sdp->use_192_bytes_for_3f)
2727 first_len = 192;
2728 dbd = 0;
2730 } else if (sdp->type == TYPE_RBC) {
2731 modepage = 6;
2732 dbd = 8;
2733 } else {
2734 modepage = 8;
2735 dbd = 0;
2738 /* cautiously ask */
2739 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2740 &data, &sshdr);
2742 if (!scsi_status_is_good(res))
2743 goto bad_sense;
2745 if (!data.header_length) {
2746 modepage = 6;
2747 first_len = 0;
2748 sd_first_printk(KERN_ERR, sdkp,
2749 "Missing header in MODE_SENSE response\n");
2752 /* that went OK, now ask for the proper length */
2753 len = data.length;
2756 * We're only interested in the first three bytes, actually.
2757 * But the data cache page is defined for the first 20.
2759 if (len < 3)
2760 goto bad_sense;
2761 else if (len > SD_BUF_SIZE) {
2762 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2763 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2764 len = SD_BUF_SIZE;
2766 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2767 len = 192;
2769 /* Get the data */
2770 if (len > first_len)
2771 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2772 &data, &sshdr);
2774 if (scsi_status_is_good(res)) {
2775 int offset = data.header_length + data.block_descriptor_length;
2777 while (offset < len) {
2778 u8 page_code = buffer[offset] & 0x3F;
2779 u8 spf = buffer[offset] & 0x40;
2781 if (page_code == 8 || page_code == 6) {
2782 /* We're interested only in the first 3 bytes.
2784 if (len - offset <= 2) {
2785 sd_first_printk(KERN_ERR, sdkp,
2786 "Incomplete mode parameter "
2787 "data\n");
2788 goto defaults;
2789 } else {
2790 modepage = page_code;
2791 goto Page_found;
2793 } else {
2794 /* Go to the next page */
2795 if (spf && len - offset > 3)
2796 offset += 4 + (buffer[offset+2] << 8) +
2797 buffer[offset+3];
2798 else if (!spf && len - offset > 1)
2799 offset += 2 + buffer[offset+1];
2800 else {
2801 sd_first_printk(KERN_ERR, sdkp,
2802 "Incomplete mode "
2803 "parameter data\n");
2804 goto defaults;
2809 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2810 goto defaults;
2812 Page_found:
2813 if (modepage == 8) {
2814 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2815 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2816 } else {
2817 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2818 sdkp->RCD = 0;
2821 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2822 if (sdp->broken_fua) {
2823 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2824 sdkp->DPOFUA = 0;
2825 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2826 !sdkp->device->use_16_for_rw) {
2827 sd_first_printk(KERN_NOTICE, sdkp,
2828 "Uses READ/WRITE(6), disabling FUA\n");
2829 sdkp->DPOFUA = 0;
2832 /* No cache flush allowed for write protected devices */
2833 if (sdkp->WCE && sdkp->write_prot)
2834 sdkp->WCE = 0;
2836 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2837 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2838 sd_printk(KERN_NOTICE, sdkp,
2839 "Write cache: %s, read cache: %s, %s\n",
2840 sdkp->WCE ? "enabled" : "disabled",
2841 sdkp->RCD ? "disabled" : "enabled",
2842 sdkp->DPOFUA ? "supports DPO and FUA"
2843 : "doesn't support DPO or FUA");
2845 return;
2848 bad_sense:
2849 if (scsi_sense_valid(&sshdr) &&
2850 sshdr.sense_key == ILLEGAL_REQUEST &&
2851 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2852 /* Invalid field in CDB */
2853 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2854 else
2855 sd_first_printk(KERN_ERR, sdkp,
2856 "Asking for cache data failed\n");
2858 defaults:
2859 if (sdp->wce_default_on) {
2860 sd_first_printk(KERN_NOTICE, sdkp,
2861 "Assuming drive cache: write back\n");
2862 sdkp->WCE = 1;
2863 } else {
2864 sd_first_printk(KERN_ERR, sdkp,
2865 "Assuming drive cache: write through\n");
2866 sdkp->WCE = 0;
2868 sdkp->RCD = 0;
2869 sdkp->DPOFUA = 0;
2873 * The ATO bit indicates whether the DIF application tag is available
2874 * for use by the operating system.
2876 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2878 int res, offset;
2879 struct scsi_device *sdp = sdkp->device;
2880 struct scsi_mode_data data;
2881 struct scsi_sense_hdr sshdr;
2883 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2884 return;
2886 if (sdkp->protection_type == 0)
2887 return;
2889 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2890 sdkp->max_retries, &data, &sshdr);
2892 if (!scsi_status_is_good(res) || !data.header_length ||
2893 data.length < 6) {
2894 sd_first_printk(KERN_WARNING, sdkp,
2895 "getting Control mode page failed, assume no ATO\n");
2897 if (scsi_sense_valid(&sshdr))
2898 sd_print_sense_hdr(sdkp, &sshdr);
2900 return;
2903 offset = data.header_length + data.block_descriptor_length;
2905 if ((buffer[offset] & 0x3f) != 0x0a) {
2906 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2907 return;
2910 if ((buffer[offset + 5] & 0x80) == 0)
2911 return;
2913 sdkp->ATO = 1;
2915 return;
2919 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2920 * @sdkp: disk to query
2922 static void sd_read_block_limits(struct scsi_disk *sdkp)
2924 unsigned int sector_sz = sdkp->device->sector_size;
2925 const int vpd_len = 64;
2926 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2928 if (!buffer ||
2929 /* Block Limits VPD */
2930 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2931 goto out;
2933 blk_queue_io_min(sdkp->disk->queue,
2934 get_unaligned_be16(&buffer[6]) * sector_sz);
2936 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2937 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2939 if (buffer[3] == 0x3c) {
2940 unsigned int lba_count, desc_count;
2942 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2944 if (!sdkp->lbpme)
2945 goto out;
2947 lba_count = get_unaligned_be32(&buffer[20]);
2948 desc_count = get_unaligned_be32(&buffer[24]);
2950 if (lba_count && desc_count)
2951 sdkp->max_unmap_blocks = lba_count;
2953 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2955 if (buffer[32] & 0x80)
2956 sdkp->unmap_alignment =
2957 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2959 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2961 if (sdkp->max_unmap_blocks)
2962 sd_config_discard(sdkp, SD_LBP_UNMAP);
2963 else
2964 sd_config_discard(sdkp, SD_LBP_WS16);
2966 } else { /* LBP VPD page tells us what to use */
2967 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2968 sd_config_discard(sdkp, SD_LBP_UNMAP);
2969 else if (sdkp->lbpws)
2970 sd_config_discard(sdkp, SD_LBP_WS16);
2971 else if (sdkp->lbpws10)
2972 sd_config_discard(sdkp, SD_LBP_WS10);
2973 else
2974 sd_config_discard(sdkp, SD_LBP_DISABLE);
2978 out:
2979 kfree(buffer);
2983 * sd_read_block_characteristics - Query block dev. characteristics
2984 * @sdkp: disk to query
2986 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2988 struct request_queue *q = sdkp->disk->queue;
2989 unsigned char *buffer;
2990 u16 rot;
2991 const int vpd_len = 64;
2993 buffer = kmalloc(vpd_len, GFP_KERNEL);
2995 if (!buffer ||
2996 /* Block Device Characteristics VPD */
2997 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2998 goto out;
3000 rot = get_unaligned_be16(&buffer[4]);
3002 if (rot == 1) {
3003 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3004 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3007 if (sdkp->device->type == TYPE_ZBC) {
3008 /* Host-managed */
3009 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
3010 } else {
3011 sdkp->zoned = (buffer[8] >> 4) & 3;
3012 if (sdkp->zoned == 1) {
3013 /* Host-aware */
3014 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
3015 } else {
3016 /* Regular disk or drive managed disk */
3017 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
3021 if (!sdkp->first_scan)
3022 goto out;
3024 if (blk_queue_is_zoned(q)) {
3025 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
3026 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
3027 } else {
3028 if (sdkp->zoned == 1)
3029 sd_printk(KERN_NOTICE, sdkp,
3030 "Host-aware SMR disk used as regular disk\n");
3031 else if (sdkp->zoned == 2)
3032 sd_printk(KERN_NOTICE, sdkp,
3033 "Drive-managed SMR disk\n");
3036 out:
3037 kfree(buffer);
3041 * sd_read_block_provisioning - Query provisioning VPD page
3042 * @sdkp: disk to query
3044 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3046 unsigned char *buffer;
3047 const int vpd_len = 8;
3049 if (sdkp->lbpme == 0)
3050 return;
3052 buffer = kmalloc(vpd_len, GFP_KERNEL);
3054 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3055 goto out;
3057 sdkp->lbpvpd = 1;
3058 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3059 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3060 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3062 out:
3063 kfree(buffer);
3066 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3068 struct scsi_device *sdev = sdkp->device;
3070 if (sdev->host->no_write_same) {
3071 sdev->no_write_same = 1;
3073 return;
3076 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3077 /* too large values might cause issues with arcmsr */
3078 int vpd_buf_len = 64;
3080 sdev->no_report_opcodes = 1;
3082 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3083 * CODES is unsupported and the device has an ATA
3084 * Information VPD page (SAT).
3086 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3087 sdev->no_write_same = 1;
3090 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3091 sdkp->ws16 = 1;
3093 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3094 sdkp->ws10 = 1;
3097 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3099 struct scsi_device *sdev = sdkp->device;
3101 if (!sdev->security_supported)
3102 return;
3104 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3105 SECURITY_PROTOCOL_IN) == 1 &&
3106 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3107 SECURITY_PROTOCOL_OUT) == 1)
3108 sdkp->security = 1;
3112 * Determine the device's preferred I/O size for reads and writes
3113 * unless the reported value is unreasonably small, large, not a
3114 * multiple of the physical block size, or simply garbage.
3116 static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3117 unsigned int dev_max)
3119 struct scsi_device *sdp = sdkp->device;
3120 unsigned int opt_xfer_bytes =
3121 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3123 if (sdkp->opt_xfer_blocks == 0)
3124 return false;
3126 if (sdkp->opt_xfer_blocks > dev_max) {
3127 sd_first_printk(KERN_WARNING, sdkp,
3128 "Optimal transfer size %u logical blocks " \
3129 "> dev_max (%u logical blocks)\n",
3130 sdkp->opt_xfer_blocks, dev_max);
3131 return false;
3134 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3135 sd_first_printk(KERN_WARNING, sdkp,
3136 "Optimal transfer size %u logical blocks " \
3137 "> sd driver limit (%u logical blocks)\n",
3138 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3139 return false;
3142 if (opt_xfer_bytes < PAGE_SIZE) {
3143 sd_first_printk(KERN_WARNING, sdkp,
3144 "Optimal transfer size %u bytes < " \
3145 "PAGE_SIZE (%u bytes)\n",
3146 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3147 return false;
3150 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3151 sd_first_printk(KERN_WARNING, sdkp,
3152 "Optimal transfer size %u bytes not a " \
3153 "multiple of physical block size (%u bytes)\n",
3154 opt_xfer_bytes, sdkp->physical_block_size);
3155 return false;
3158 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3159 opt_xfer_bytes);
3160 return true;
3164 * sd_revalidate_disk - called the first time a new disk is seen,
3165 * performs disk spin up, read_capacity, etc.
3166 * @disk: struct gendisk we care about
3168 static int sd_revalidate_disk(struct gendisk *disk)
3170 struct scsi_disk *sdkp = scsi_disk(disk);
3171 struct scsi_device *sdp = sdkp->device;
3172 struct request_queue *q = sdkp->disk->queue;
3173 sector_t old_capacity = sdkp->capacity;
3174 unsigned char *buffer;
3175 unsigned int dev_max, rw_max;
3177 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3178 "sd_revalidate_disk\n"));
3181 * If the device is offline, don't try and read capacity or any
3182 * of the other niceties.
3184 if (!scsi_device_online(sdp))
3185 goto out;
3187 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3188 if (!buffer) {
3189 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3190 "allocation failure.\n");
3191 goto out;
3194 sd_spinup_disk(sdkp);
3197 * Without media there is no reason to ask; moreover, some devices
3198 * react badly if we do.
3200 if (sdkp->media_present) {
3201 sd_read_capacity(sdkp, buffer);
3204 * set the default to rotational. All non-rotational devices
3205 * support the block characteristics VPD page, which will
3206 * cause this to be updated correctly and any device which
3207 * doesn't support it should be treated as rotational.
3209 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3210 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3212 if (scsi_device_supports_vpd(sdp)) {
3213 sd_read_block_provisioning(sdkp);
3214 sd_read_block_limits(sdkp);
3215 sd_read_block_characteristics(sdkp);
3216 sd_zbc_read_zones(sdkp, buffer);
3219 sd_print_capacity(sdkp, old_capacity);
3221 sd_read_write_protect_flag(sdkp, buffer);
3222 sd_read_cache_type(sdkp, buffer);
3223 sd_read_app_tag_own(sdkp, buffer);
3224 sd_read_write_same(sdkp, buffer);
3225 sd_read_security(sdkp, buffer);
3229 * We now have all cache related info, determine how we deal
3230 * with flush requests.
3232 sd_set_flush_flag(sdkp);
3234 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3235 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3237 /* Some devices report a maximum block count for READ/WRITE requests. */
3238 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3239 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3241 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3242 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3243 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3244 } else {
3245 q->limits.io_opt = 0;
3246 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3247 (sector_t)BLK_DEF_MAX_SECTORS);
3250 /* Do not exceed controller limit */
3251 rw_max = min(rw_max, queue_max_hw_sectors(q));
3254 * Only update max_sectors if previously unset or if the current value
3255 * exceeds the capabilities of the hardware.
3257 if (sdkp->first_scan ||
3258 q->limits.max_sectors > q->limits.max_dev_sectors ||
3259 q->limits.max_sectors > q->limits.max_hw_sectors)
3260 q->limits.max_sectors = rw_max;
3262 sdkp->first_scan = 0;
3264 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3265 sd_config_write_same(sdkp);
3266 kfree(buffer);
3269 * For a zoned drive, revalidating the zones can be done only once
3270 * the gendisk capacity is set. So if this fails, set back the gendisk
3271 * capacity to 0.
3273 if (sd_zbc_revalidate_zones(sdkp))
3274 set_capacity_and_notify(disk, 0);
3276 out:
3277 return 0;
3281 * sd_unlock_native_capacity - unlock native capacity
3282 * @disk: struct gendisk to set capacity for
3284 * Block layer calls this function if it detects that partitions
3285 * on @disk reach beyond the end of the device. If the SCSI host
3286 * implements ->unlock_native_capacity() method, it's invoked to
3287 * give it a chance to adjust the device capacity.
3289 * CONTEXT:
3290 * Defined by block layer. Might sleep.
3292 static void sd_unlock_native_capacity(struct gendisk *disk)
3294 struct scsi_device *sdev = scsi_disk(disk)->device;
3296 if (sdev->host->hostt->unlock_native_capacity)
3297 sdev->host->hostt->unlock_native_capacity(sdev);
3301 * sd_format_disk_name - format disk name
3302 * @prefix: name prefix - ie. "sd" for SCSI disks
3303 * @index: index of the disk to format name for
3304 * @buf: output buffer
3305 * @buflen: length of the output buffer
3307 * SCSI disk names starts at sda. The 26th device is sdz and the
3308 * 27th is sdaa. The last one for two lettered suffix is sdzz
3309 * which is followed by sdaaa.
3311 * This is basically 26 base counting with one extra 'nil' entry
3312 * at the beginning from the second digit on and can be
3313 * determined using similar method as 26 base conversion with the
3314 * index shifted -1 after each digit is computed.
3316 * CONTEXT:
3317 * Don't care.
3319 * RETURNS:
3320 * 0 on success, -errno on failure.
3322 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3324 const int base = 'z' - 'a' + 1;
3325 char *begin = buf + strlen(prefix);
3326 char *end = buf + buflen;
3327 char *p;
3328 int unit;
3330 p = end - 1;
3331 *p = '\0';
3332 unit = base;
3333 do {
3334 if (p == begin)
3335 return -EINVAL;
3336 *--p = 'a' + (index % unit);
3337 index = (index / unit) - 1;
3338 } while (index >= 0);
3340 memmove(begin, p, end - p);
3341 memcpy(buf, prefix, strlen(prefix));
3343 return 0;
3347 * sd_probe - called during driver initialization and whenever a
3348 * new scsi device is attached to the system. It is called once
3349 * for each scsi device (not just disks) present.
3350 * @dev: pointer to device object
3352 * Returns 0 if successful (or not interested in this scsi device
3353 * (e.g. scanner)); 1 when there is an error.
3355 * Note: this function is invoked from the scsi mid-level.
3356 * This function sets up the mapping between a given
3357 * <host,channel,id,lun> (found in sdp) and new device name
3358 * (e.g. /dev/sda). More precisely it is the block device major
3359 * and minor number that is chosen here.
3361 * Assume sd_probe is not re-entrant (for time being)
3362 * Also think about sd_probe() and sd_remove() running coincidentally.
3364 static int sd_probe(struct device *dev)
3366 struct scsi_device *sdp = to_scsi_device(dev);
3367 struct scsi_disk *sdkp;
3368 struct gendisk *gd;
3369 int index;
3370 int error;
3372 scsi_autopm_get_device(sdp);
3373 error = -ENODEV;
3374 if (sdp->type != TYPE_DISK &&
3375 sdp->type != TYPE_ZBC &&
3376 sdp->type != TYPE_MOD &&
3377 sdp->type != TYPE_RBC)
3378 goto out;
3380 #ifndef CONFIG_BLK_DEV_ZONED
3381 if (sdp->type == TYPE_ZBC)
3382 goto out;
3383 #endif
3384 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3385 "sd_probe\n"));
3387 error = -ENOMEM;
3388 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3389 if (!sdkp)
3390 goto out;
3392 gd = alloc_disk(SD_MINORS);
3393 if (!gd)
3394 goto out_free;
3396 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3397 if (index < 0) {
3398 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3399 goto out_put;
3402 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3403 if (error) {
3404 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3405 goto out_free_index;
3408 sdkp->device = sdp;
3409 sdkp->driver = &sd_template;
3410 sdkp->disk = gd;
3411 sdkp->index = index;
3412 sdkp->max_retries = SD_MAX_RETRIES;
3413 atomic_set(&sdkp->openers, 0);
3414 atomic_set(&sdkp->device->ioerr_cnt, 0);
3416 if (!sdp->request_queue->rq_timeout) {
3417 if (sdp->type != TYPE_MOD)
3418 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3419 else
3420 blk_queue_rq_timeout(sdp->request_queue,
3421 SD_MOD_TIMEOUT);
3424 device_initialize(&sdkp->dev);
3425 sdkp->dev.parent = dev;
3426 sdkp->dev.class = &sd_disk_class;
3427 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3429 error = device_add(&sdkp->dev);
3430 if (error)
3431 goto out_free_index;
3433 get_device(dev);
3434 dev_set_drvdata(dev, sdkp);
3436 gd->major = sd_major((index & 0xf0) >> 4);
3437 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3439 gd->fops = &sd_fops;
3440 gd->private_data = &sdkp->driver;
3441 gd->queue = sdkp->device->request_queue;
3443 /* defaults, until the device tells us otherwise */
3444 sdp->sector_size = 512;
3445 sdkp->capacity = 0;
3446 sdkp->media_present = 1;
3447 sdkp->write_prot = 0;
3448 sdkp->cache_override = 0;
3449 sdkp->WCE = 0;
3450 sdkp->RCD = 0;
3451 sdkp->ATO = 0;
3452 sdkp->first_scan = 1;
3453 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3455 sd_revalidate_disk(gd);
3457 gd->flags = GENHD_FL_EXT_DEVT;
3458 if (sdp->removable) {
3459 gd->flags |= GENHD_FL_REMOVABLE;
3460 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3461 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3464 blk_pm_runtime_init(sdp->request_queue, dev);
3465 if (sdp->rpm_autosuspend) {
3466 pm_runtime_set_autosuspend_delay(dev,
3467 sdp->host->hostt->rpm_autosuspend_delay);
3469 device_add_disk(dev, gd, NULL);
3470 if (sdkp->capacity)
3471 sd_dif_config_host(sdkp);
3473 sd_revalidate_disk(gd);
3475 if (sdkp->security) {
3476 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3477 if (sdkp->opal_dev)
3478 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3481 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3482 sdp->removable ? "removable " : "");
3483 scsi_autopm_put_device(sdp);
3485 return 0;
3487 out_free_index:
3488 ida_free(&sd_index_ida, index);
3489 out_put:
3490 put_disk(gd);
3491 out_free:
3492 sd_zbc_release_disk(sdkp);
3493 kfree(sdkp);
3494 out:
3495 scsi_autopm_put_device(sdp);
3496 return error;
3500 * sd_remove - called whenever a scsi disk (previously recognized by
3501 * sd_probe) is detached from the system. It is called (potentially
3502 * multiple times) during sd module unload.
3503 * @dev: pointer to device object
3505 * Note: this function is invoked from the scsi mid-level.
3506 * This function potentially frees up a device name (e.g. /dev/sdc)
3507 * that could be re-used by a subsequent sd_probe().
3508 * This function is not called when the built-in sd driver is "exit-ed".
3510 static int sd_remove(struct device *dev)
3512 struct scsi_disk *sdkp;
3513 dev_t devt;
3515 sdkp = dev_get_drvdata(dev);
3516 devt = disk_devt(sdkp->disk);
3517 scsi_autopm_get_device(sdkp->device);
3519 async_synchronize_full_domain(&scsi_sd_pm_domain);
3520 device_del(&sdkp->dev);
3521 del_gendisk(sdkp->disk);
3522 sd_shutdown(dev);
3524 free_opal_dev(sdkp->opal_dev);
3526 mutex_lock(&sd_ref_mutex);
3527 dev_set_drvdata(dev, NULL);
3528 put_device(&sdkp->dev);
3529 mutex_unlock(&sd_ref_mutex);
3531 return 0;
3535 * scsi_disk_release - Called to free the scsi_disk structure
3536 * @dev: pointer to embedded class device
3538 * sd_ref_mutex must be held entering this routine. Because it is
3539 * called on last put, you should always use the scsi_disk_get()
3540 * scsi_disk_put() helpers which manipulate the semaphore directly
3541 * and never do a direct put_device.
3543 static void scsi_disk_release(struct device *dev)
3545 struct scsi_disk *sdkp = to_scsi_disk(dev);
3546 struct gendisk *disk = sdkp->disk;
3547 struct request_queue *q = disk->queue;
3549 ida_free(&sd_index_ida, sdkp->index);
3552 * Wait until all requests that are in progress have completed.
3553 * This is necessary to avoid that e.g. scsi_end_request() crashes
3554 * due to clearing the disk->private_data pointer. Wait from inside
3555 * scsi_disk_release() instead of from sd_release() to avoid that
3556 * freezing and unfreezing the request queue affects user space I/O
3557 * in case multiple processes open a /dev/sd... node concurrently.
3559 blk_mq_freeze_queue(q);
3560 blk_mq_unfreeze_queue(q);
3562 disk->private_data = NULL;
3563 put_disk(disk);
3564 put_device(&sdkp->device->sdev_gendev);
3566 sd_zbc_release_disk(sdkp);
3568 kfree(sdkp);
3571 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3573 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3574 struct scsi_sense_hdr sshdr;
3575 struct scsi_device *sdp = sdkp->device;
3576 int res;
3578 if (start)
3579 cmd[4] |= 1; /* START */
3581 if (sdp->start_stop_pwr_cond)
3582 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3584 if (!scsi_device_online(sdp))
3585 return -ENODEV;
3587 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3588 SD_TIMEOUT, sdkp->max_retries, 0, RQF_PM, NULL);
3589 if (res) {
3590 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3591 if (driver_byte(res) == DRIVER_SENSE)
3592 sd_print_sense_hdr(sdkp, &sshdr);
3593 if (scsi_sense_valid(&sshdr) &&
3594 /* 0x3a is medium not present */
3595 sshdr.asc == 0x3a)
3596 res = 0;
3599 /* SCSI error codes must not go to the generic layer */
3600 if (res)
3601 return -EIO;
3603 return 0;
3607 * Send a SYNCHRONIZE CACHE instruction down to the device through
3608 * the normal SCSI command structure. Wait for the command to
3609 * complete.
3611 static void sd_shutdown(struct device *dev)
3613 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3615 if (!sdkp)
3616 return; /* this can happen */
3618 if (pm_runtime_suspended(dev))
3619 return;
3621 if (sdkp->WCE && sdkp->media_present) {
3622 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3623 sd_sync_cache(sdkp, NULL);
3626 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3627 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3628 sd_start_stop_device(sdkp, 0);
3632 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3634 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3635 struct scsi_sense_hdr sshdr;
3636 int ret = 0;
3638 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3639 return 0;
3641 if (sdkp->WCE && sdkp->media_present) {
3642 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3643 ret = sd_sync_cache(sdkp, &sshdr);
3645 if (ret) {
3646 /* ignore OFFLINE device */
3647 if (ret == -ENODEV)
3648 return 0;
3650 if (!scsi_sense_valid(&sshdr) ||
3651 sshdr.sense_key != ILLEGAL_REQUEST)
3652 return ret;
3655 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3656 * doesn't support sync. There's not much to do and
3657 * suspend shouldn't fail.
3659 ret = 0;
3663 if (sdkp->device->manage_start_stop) {
3664 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3665 /* an error is not worth aborting a system sleep */
3666 ret = sd_start_stop_device(sdkp, 0);
3667 if (ignore_stop_errors)
3668 ret = 0;
3671 return ret;
3674 static int sd_suspend_system(struct device *dev)
3676 return sd_suspend_common(dev, true);
3679 static int sd_suspend_runtime(struct device *dev)
3681 return sd_suspend_common(dev, false);
3684 static int sd_resume(struct device *dev)
3686 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3687 int ret;
3689 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3690 return 0;
3692 if (!sdkp->device->manage_start_stop)
3693 return 0;
3695 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3696 ret = sd_start_stop_device(sdkp, 1);
3697 if (!ret)
3698 opal_unlock_from_suspend(sdkp->opal_dev);
3699 return ret;
3703 * init_sd - entry point for this driver (both when built in or when
3704 * a module).
3706 * Note: this function registers this driver with the scsi mid-level.
3708 static int __init init_sd(void)
3710 int majors = 0, i, err;
3712 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3714 for (i = 0; i < SD_MAJORS; i++) {
3715 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
3716 continue;
3717 majors++;
3720 if (!majors)
3721 return -ENODEV;
3723 err = class_register(&sd_disk_class);
3724 if (err)
3725 goto err_out;
3727 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3728 0, 0, NULL);
3729 if (!sd_cdb_cache) {
3730 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3731 err = -ENOMEM;
3732 goto err_out_class;
3735 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3736 if (!sd_cdb_pool) {
3737 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3738 err = -ENOMEM;
3739 goto err_out_cache;
3742 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3743 if (!sd_page_pool) {
3744 printk(KERN_ERR "sd: can't init discard page pool\n");
3745 err = -ENOMEM;
3746 goto err_out_ppool;
3749 err = scsi_register_driver(&sd_template.gendrv);
3750 if (err)
3751 goto err_out_driver;
3753 return 0;
3755 err_out_driver:
3756 mempool_destroy(sd_page_pool);
3758 err_out_ppool:
3759 mempool_destroy(sd_cdb_pool);
3761 err_out_cache:
3762 kmem_cache_destroy(sd_cdb_cache);
3764 err_out_class:
3765 class_unregister(&sd_disk_class);
3766 err_out:
3767 for (i = 0; i < SD_MAJORS; i++)
3768 unregister_blkdev(sd_major(i), "sd");
3769 return err;
3773 * exit_sd - exit point for this driver (when it is a module).
3775 * Note: this function unregisters this driver from the scsi mid-level.
3777 static void __exit exit_sd(void)
3779 int i;
3781 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3783 scsi_unregister_driver(&sd_template.gendrv);
3784 mempool_destroy(sd_cdb_pool);
3785 mempool_destroy(sd_page_pool);
3786 kmem_cache_destroy(sd_cdb_cache);
3788 class_unregister(&sd_disk_class);
3790 for (i = 0; i < SD_MAJORS; i++)
3791 unregister_blkdev(sd_major(i), "sd");
3794 module_init(init_sd);
3795 module_exit(exit_sd);
3797 void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3799 scsi_print_sense_hdr(sdkp->device,
3800 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3803 void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3805 const char *hb_string = scsi_hostbyte_string(result);
3806 const char *db_string = scsi_driverbyte_string(result);
3808 if (hb_string || db_string)
3809 sd_printk(KERN_INFO, sdkp,
3810 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3811 hb_string ? hb_string : "invalid",
3812 db_string ? db_string : "invalid");
3813 else
3814 sd_printk(KERN_INFO, sdkp,
3815 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3816 msg, host_byte(result), driver_byte(result));